What is backtesting? How to read one before you trust it

Backtesting means running a trading rule over historical prices to see what it would have done: every entry, every exit, the hit rate, the equity curve and the worst drawdown. Every serious algo platform in India publishes one. Most retail traders read the return at the bottom and stop. This guide is about reading the rest.

What a backtest actually shows

A list of hypothetical trades under a fixed rule, on the data available today, usually before brokerage, taxes and slippage. From that list come the statistics: number of trades, hit rate, average win and loss, profit factor, maximum drawdown and time to recover. The equity curve is a picture of the same list. None of it is a forecast. It is a description of one path through the past.

Five ways a backtest lies

Overfitting. A rule with enough parameters can be tuned to fit any history. Ask how many parameters, and whether the result holds on data the rule was not tuned on (out-of-sample). Survivorship bias. Testing only on stocks that exist today ignores the ones that fell out of the index or delisted — the losers. Firefly's backtests retain delisted names for that reason. Look-ahead bias. Using a day's close to decide a trade placed at that day's open, or a quarterly result before it was published. Costs. A rule that trades 1,500 times a month lives or dies on brokerage, STT and slippage; a backtest "before costs" can flip sign after them. Regime. A test from 2020 to 2024 has seen one kind of market. Ask for 2011, 2013, 2018 and the first quarter of 2020.

How to read one in five minutes

• Find the worst drawdown and its date. Could you have held through it with real money? That is the real question.
• Find the longest flat stretch. Months with no new equity high. Most people quit in month four.
• Check the trade count. Under 100 trades is anecdote; statistics start in the hundreds.
• Compare in-sample and out-of-sample. If the platform cannot tell you which is which, the whole thing is in-sample.
• Look for the cost line. Brokerage, STT, slippage assumption. Missing means zero was assumed.

How Firefly publishes its backtests

Trade by trade, per strategy family, with entries, exits, stop hits, hit rate and maximum drawdown; NSE from 2010, US from 2005, crypto from 2017; delisted names retained; labelled simulated and before costs. The losing stretches are in the same tables as the winning ones. The daily output of the same rules is public on Signals, which is the out-of-sample test anyone can watch — and paper trading is the step between reading and risking.

Questions people ask

Fintrens Technologies Pvt Ltd is not a SEBI-registered investment adviser or research analyst. This guide is general information, not advice; trading involves risk of loss.